• 0 Posts
  • 62 Comments
Joined 3 years ago
cake
Cake day: November 9th, 2022

help-circle
  • 3 students share an apartment and 2 of them study a lot but the third spends most his nights partying. The 2 studious housemates decide to pull a prank on him, and one night when he comes home they are waiting for him next to the bedroom door wearing white sheets. One of the friends says ‘welcome friend, I am Peter!’. The other says ‘welcome friend, I am Paul’. The drunk house mate looks at them and says ‘Colleagues! would you mind stepping aside? I am Lazarus!’

    yeah, that doesn’t translate… in Dutch, the names refer to St peter and St Paul and both end in -us as well: Petrus and Paulus. Also, ‘being Lazarus’ means being very drunk.


  • I’m not arguing that AI won’t get better, I’m arguing that the exponential improvements in AI that op was expecting are mostly wishful thinking.

    they could stick to old data only, but then how do you keep growing the dataset by the amounts that have been done recently? that is where a lot of the (diminishing) improvements the last years have come from.

    and it is not at all clear how to apply reinforcement learning for more generic tasks like chatbots, without a clear scoring system like both chess and StarCraft have.


  • the problem is that ai’s are trained on programs that humans have written. At best the llm architectures it creates will be similar to the state of the art that humans have created at that point.

    however, even more important than the architecture of an ai model is the training data that it is trained on. If we start including ai-generated programs in this data, we will quickly observe model collapse: performance of models tend to get worse as more ai-generated data is included in the training data.

    rather than AIs generating ever smarter new AIs, the more likely result is that we can’t scrape new quality datasets as they’ve all been contaminated with llm-generated data that will only reduce model performance


  • why is it very likely to do that? we have no evidence to believe this is true at all and several decades of slow, plodding ai research that suggests real improvement comes incrementally like in other research areas.

    to me, your suggestion sounds like the result of the logical leaps made by yudkovsky and the people on his forums
















  • most of his comments I completely understand, but saying process manager ‘looks a bit low-rent’ tells me absolutely nothing other than that you personally don’t like the look.

    also, complaining about not having a consistent way to install apps is a bit rich when coming from a desktop where the best way is still to just find the app website, download the installer and click through an inconsistent installer that might be trying to add ad- or spyware along with the app you want